Cognitive balance theory was devised by Heider (1946, 1958) to explain how people resolve inconsistencies in their interpersonal affects. For example, if a person p likes another person o, and o likes object x, then p will tend to like x also. In general, such triadic relations are balanced when all three relations are positive or two of the three relations are negative. Such elementary affect triads were modeled by Read and Miller (1994) in a preliminary way.
Heider's theory was broadened and deepened by Rosenberg and Abelson (1960) who extended it beyond affect relations and added a least-effort principle, predicting that people achieve balance by changing the fewest possible cognitions. The least-effort hypothesis was supported in their role-playing experiment in which participants were told about an unbalanced situation (differently unbalanced in different conditions) and then received communications that, if accepted, could balance the situation, but only after varying numbers of cognition changes. As predicted, participants accepted the communication that achieved balance in one cognition change over those that required more cognition changes. The best accepted communication was different in each condition.
This project involves building neural-network simulations of the Rosenberg and Abelson balance phenomena. Simulations will subject their psychological hypotheses to more precise tests and perhaps generate predictions not yet tested in psychological research. The chances of being able to go beyond classical verbal theories are enhanced by the fact that neural modeling specifies reasoning mechanisms at a deeper level and includes more constraints. The basic theoretical premise governing the simulations is that balance seeking is a special cases of the tendency for people to seek consistency among their cognitions.
One simulation technique that has had considerable success in covering cognitive consistency phenomena is constraint satisfaction (Shultz & Lepper, 1996). The motivation to achieve cognitive consistency can be viewed as imposing constraints on the cognitions that an individual holds at any given moment. Consistency problems can be solved by satisfying these multiple, soft constraints. The constraints are considered to be soft because it is desirable, but not essential, to satisfy them and they may vary in their relative importance to the individual.
Simulation of balance phenomena requires that connection weights, as well as unit activations, can change in order to satisfy constraints. Unlike many consistency experiments that preserve implicational relations but allow some attitudes to change, much of the interesting balance work (e.g., Rosenberg & Abelson, 1960) involved constant attitudes and beliefs, but varying implicational relations. Past work with constraint-satisfaction networks has either kept connection weights constant and allowed unit activations to change, or kept unit activations constant and allowed connection weights to change, as in Hebbian learning. It is not at all clear that allowing both connection weights and unit activations to change simultaneously will enable networks to converge in a stable fashion to achieve a high degree of balance. Perhaps we can do this by making connection weights or unit activations more or less resistant to change. The basic idea is that the presence of sufficiently resistant items (connection weights and/or unit activations) may allow stable convergence to balanced states.
Sufficient facility in Lisp to use a neural-network simulator. Ability to read social-psychology experiments and to use statistical packages such as SPSS for ANOVA.
Heider, F. (1946). Attitudes and cognitive organization. Journal of Personality, 21, 107-112.
Heider, F. (1958). The
psychology of interpersonal relations.
Read, S. J., &
Miller, L. C. (1994). Dissonance and balance in belief systems: The promise of
parallel constraint satisfaction processes and connectionist modeling
approaches. In R. C. Schank & E. Langer (Eds.), Beliefs, reasoning, and
decision making: Psycho-logic in honor of Bob Abelson (pp. 209-235).
Rosenberg, M. J.,
& Abelson, R. P. (1960). An analysis of cognitive balancing. In M. J.
Rosenberg, C. I. Hovland, W. J. McGuire, R. P. Abelson, & J. W. Brehm
(Eds.), Attitude organization and change (pp. 112-163).
Shultz, T. R., & Lepper, M. R. (1996). Cognitive dissonance reduction as constraint satisfaction. Psychological Review, 103, 219-240.